Dissertation Title:  Human-AI Interaction for Human Connection 

Abstract:

Humans rarely navigate the world in isolation -- through social interaction and communication, we share who we are with the world, and the world becomes part of us. Current artificially intelligent systems have the power to tailor our social understanding of others and digital technologies allow us to communicate across both space and time, yet, loneliness and apathy are widespread. This dissertation explores how human-AI interaction with socially intelligent systems impacts authentic human connection. To this end, I argue that understanding AI's role in connection requires studying it as a participant in the social ecology across individual, dyadic, and collective levels of analysis: (1) What relational information must AI encode to understand human sociality (representation)? (2) Can AI participate in interactions that improve interpersonal connection and social communication (mediation)? (3) And finally, what risks to social agency emerge when AI-mediated interactions scale to networks and populations (amplification)?

In Part I of this thesis, I develop computational methods to model and evaluate social behaviors--such as empathy and conflict mediation--enabling AI systems that encode and reason about relational information. Here, I lay the groundwork for what content humans empathize with, how communication style shapes emotional reactions, and why social contexts and personal experiences impact social perceptions.

In Part II, I evaluate how AI can participate in interpersonal dynamics, through design and real-world evaluation of interactive interfaces that foster empathetic story matching and social-emotional learning. Finally, in Part III, I examine how AI can shape human social worlds at scale, potentially exacerbating social harms like emotional manipulation, and establishing mitigation strategies to protect users from nefarious social influence.

In summary, this dissertation contributes computational modeling of relational information, establishing methods and benchmarks for how AI encodes socially-relevant knowledge, human-AI interaction paradigms for mediating social relationships, and safeguards against socio-emotional manipulation of users. As AI-mediated communication becomes increasingly embedded in daily life, this dissertation offers both theoretical and applied foundations for designing socially intelligent systems that strengthen human connection while safeguarding individuals from socio-emotional exploitation.

Committee members:

Dr. Cynthia Breazeal
Professor of Media Arts and Sciences
Massachusetts Institute of Technology

Dr. Maarten Sap
Assistant Professor of Computer Science
Carnegie Mellon University

Dr. Pattie Maes
Professor of Media Arts and Sciences
Massachusetts Institute of Technology

Dr. Hae Won Park
Research Scientist
Massachusetts Institute of Technology